This paper proposes a unified approach for bottom-up hierarchical image segmentation and object candidate generation called Multiscale Combinatorial Grouping (MCG). The approach includes a fast normalized cuts algorithm, a high-performance hierarchical segmenter that leverages multiscale information, and a grouping strategy that combines multiscale regions into accurate object candidates by exploring their combinatorial space. The method is evaluated on the BSDS500 and PASCAL 2012 segmentation datasets, showing state-of-the-art results in contour detection, hierarchical segmentation, and object candidate generation. The MCG approach produces high-quality contours, hierarchical regions, and object candidates by using a multiscale hierarchical segmentation algorithm. It first computes a multiscale segmentation hierarchy, then uses combinatorial grouping to generate object candidates. The algorithm efficiently explores the combinatorial space of multiscale regions to produce accurate object candidates. The MCG approach is evaluated on the BSDS500 dataset, where it achieves the best results in contour detection and hierarchical segmentation. On the PASCAL 2012 dataset, it achieves state-of-the-art object-level accuracy, with a relative improvement of +20% over Selective Search at 1100 candidates per image and +7.8% over CPMC at 100 candidates per image. The MCG approach is also compared with other state-of-the-art methods, showing superior performance in both instance-level and class-level metrics. It is also compared with a single-scale version of MCG (SCG), which is faster and produces comparable results. The MCG approach is efficient and scalable, with a fast eigenvector computation for normalized-cut segmentation and an efficient algorithm for combinatorial merging of hierarchical regions. The approach is publicly available, with code, results, and evaluation protocols. The method is effective for bottom-up segmentation and object candidate generation, and has the potential to be applied to a wide range of computer vision tasks.This paper proposes a unified approach for bottom-up hierarchical image segmentation and object candidate generation called Multiscale Combinatorial Grouping (MCG). The approach includes a fast normalized cuts algorithm, a high-performance hierarchical segmenter that leverages multiscale information, and a grouping strategy that combines multiscale regions into accurate object candidates by exploring their combinatorial space. The method is evaluated on the BSDS500 and PASCAL 2012 segmentation datasets, showing state-of-the-art results in contour detection, hierarchical segmentation, and object candidate generation. The MCG approach produces high-quality contours, hierarchical regions, and object candidates by using a multiscale hierarchical segmentation algorithm. It first computes a multiscale segmentation hierarchy, then uses combinatorial grouping to generate object candidates. The algorithm efficiently explores the combinatorial space of multiscale regions to produce accurate object candidates. The MCG approach is evaluated on the BSDS500 dataset, where it achieves the best results in contour detection and hierarchical segmentation. On the PASCAL 2012 dataset, it achieves state-of-the-art object-level accuracy, with a relative improvement of +20% over Selective Search at 1100 candidates per image and +7.8% over CPMC at 100 candidates per image. The MCG approach is also compared with other state-of-the-art methods, showing superior performance in both instance-level and class-level metrics. It is also compared with a single-scale version of MCG (SCG), which is faster and produces comparable results. The MCG approach is efficient and scalable, with a fast eigenvector computation for normalized-cut segmentation and an efficient algorithm for combinatorial merging of hierarchical regions. The approach is publicly available, with code, results, and evaluation protocols. The method is effective for bottom-up segmentation and object candidate generation, and has the potential to be applied to a wide range of computer vision tasks.